Kenneth H. Fielding

Neural networks for automatic target recognition

Abstract Many applications reported in artificial neural networks are associated with military problems. This paper reviews concepts associated with the processing of military data to find and recognize targets—automatic target recognition (ATR). A general-purpose automatic target recognition system does not exist. The work presented here is demonstrated on military data, but it can only be consideredproof of principle until systems are fielded andproven “under-fire”. ATR data can be in the form of non-imaging one-dimensional sensor returns, such as ultra-high range-resolution radar returns for air-to-air automatic target recognition and vibration signatures from a laser radar for recognition of ground targets. The ATR data can be two-dimensional images. The most common ATR images are infrared, but current systems must also deal with synthetic aperture radar images. Finally, the data can be three-dimensional, such as sequences of multiple exposures taken over time from a nonstationary world. Targets move, as do sensors, and that movement can be exploited by the ATR. Hyperspectral data, which are views of the same piece of the world looking at different spectral bands, is another example of multiple image data; the third dimension is now wavelength and not time. ATR system design usually consists of four stages. The first stage is to select the sensor or sensors to produce the target measurements. The next stage is the preprocessing of the data and the location of regions of interest within the data (segmentation). The human retina is a ruthless preprocessor. Physiology motivated preprocessing and segmentation is demonstrated along with supervised and unsupervised artificial neural segmentation techniques. The third design step is feature extraction and selection: the extraction of a set of numbers which characterize regions of the data. The last step is the processing of the features for decision making (classification). The area of classification is where most ATR related neural network research has been accomplished. The relation of neural classifiers to Bayesian techniques is emphasized along with the more recent use of feature sequences to enhance classification. The principal theme of this paper is that artificial neural networks have proven to be an interesting and useful alternate processing strategy. Artificial neural techniques, however, are not magical solutions with mystical abilities that work without good engineering. Good understanding of the capabilities and limitations of neural techniques is required to apply them productively to ATR problems.

1-f binary joint transform correlator

A one lens focal length binary joint transform correlator is described. This correlator uses a magneto-optic spatial light modulator, lens, and standard 8-bit resolution CCD camera. Computer simulations and experimental results of the effects of changes in scale, in-plane rotation, out-of-plane rotation, target/reference separation, and multiple targets are discussed. The performance using actual sensor imagery containing clutter is presented.

Optical fingerprint identification by binary joint transform correlation

We describe an optical fingerprint identification system that optically reads a latent fingerprint for correlation using a binary joint transform correlator. The fingerprint is read using the total internal reflection property of a prism. The system was built, tested, and the experimental results are presented.

Spatio-temporal pattern recognition using hidden Markov models

A spatio-temporal method for identifying objects contained in an image sequence is presented. The Hidden Markov Model (HMM) technique is used as the classification algorithm, making classification decisions based on a spatio-temporal sequence of observed object features. A five class problem is considered. Classification accuracies of 100% and 99.7%, are obtained for sequences of images generated over two separate regions of viewing positions. HMMs trained on image sequences of the objects moving in opposite directions showed a 98.1% successful classification rate by class and direction of movement. The HMM technique proved robust to image corruption with additive correlated noise and had a higher accuracy than a single-look nearest neighbor method. A real image sequence of one of the objects used was successfully recognized with the HMMs trained on synthetic data. This study shows the temporal changes that observed feature vectors undergo due to object motion hold information that can yield superior classification accuracy when compared with single-frame techniques.<<ETX>>

Optical Haar wavelet transform

An optical Haar mother wavelet is created with a Semetex 128 x 128 magneto-optic spatial light modulator. Two techniques for dilating the mother wavelet are explored: (1) aperture stopping and (2) operating the SLM in ternary phase-amplitude mode. Discrete resolution levels of a continuous wavelet transform are obtained by optically correlating a binarized image with multiple dilations ofthe mother wavelet. Frequency-plane masks for the correlation process are generated using thermoplastic holography. Experimental results are compared with a digital simulation of the wavelet transform.

Recognition of moving light displays using hidden Markov models

Abstract A spatio-temporal method of identifying moving light displays (M LDs) is presented. The hidden Markov model (HMM) technique is used as the classification algorithm, making classification decisions based on a spatio-temporal sequence of observed object features. Individual frames of a MLD image sequence are assumed to be segmented and contain very little spatial information. The information content is highly temporal in the sense that image sequences are required for object identification. A single look and alternate multiple frame classifier are used for comparison with the HMM technique. A three-class problem is considered. The single look average classification rate for the moving light display imagery was observed to be near 50%. In contrast, the hidden Markov model average classification rate was above 93%. The alternate nearest neighbor multiple frame technique average classification rate was 20% below the hidden Markov models. A one sided t-test revealed a highly statistically significant difference between the hidden Markov model and multiple frame technique at a 0.01 level of significance.

Clutter Effects on Optical Correlators

We examine the effect of varying ground clutter on the performance of optical correlators. Comparisons are made between the classical matched (CMF), phase-only (POF), and binary phase-only (BPOF) filter. A new definition of the signal-to-clutter ratio (SCR) in the input scene and the signal-to-noise ratio (SNR) in the correlation plane is presented. Simulations show the POF and BPOF always outperform the CMF. The BPOF performs best in scenes with low SCR ratios (high clutter) using the new metric. The POF performs best for all levels of SCR when a block is placed in the filter plane.

Spatiotemporal pattern recognition using hidden Markov models

A spatio-temporal method for identifying objects contained in an image sequence is presented. The Hidden Markov Model (HMM) technique is used as the classification algorithm, making classification decisions based on a spatio-temporal sequence of observed object features. A five class problem is considered. Classification accuracies of 100% and 99.7% are obtained for sequences of images generated over two separate regions of viewing positions. HMMs trained on image sequences of the objects moving in opposite directions showed a 98.1% successful classification rate by class and direction of movement. The HMM technique proved robust to image corruption with additive correlated noise and had a higher accuracy than a single look nearest neighbor method. A real image sequence of one of the objects used was successfully recognized with the HMMs trained on synthetic data. This study shows the temporal changes that observed feature vectors undergo due to object motion hold information that can yield superior classification accuracy when compared to single frame techniques.

An application of embedology to spatio-temporal pattern recognition

The theory of embedded time series is shown applicable for determining a reasonable lower bound on the length of test sequence required for accurate classification of moving objects. Sequentially recorded feature vectors of a moving object form a training trajectory in feature space. Each of the sequences of feature vector components is a time series, and under certain conditions, each of these time series has approximately the same fractal dimension. The embedding theorem may be applied to this fractal dimension to establish a sufficient number of observations to determine the feature space trajectory of the object. It is argued that this number is a reasonable lower bound on test sequence length for use in object classification. Experiments with data corresponding to five military vehicles (observed following a projected Lorenz trajectory on a viewing sphere) show that this bound is indeed adequate.

Position, Scale, And Rotation Invariant Holographic Associative Memory

An all-optical holographic memory was recently proposed by researchers at the Hughes Research Laboratories. This paper details the investigation into the characteristics of that system and an extension to a position, scale, and rotation invariant (PSRI) holographic associative memory. The PSRI feature space is the en-polar representation of the square magnitude of the Fourier transform, IF(fnr, fo)I2, of the objects. This representation is generated optically using a coordinate transform computer-generated hologram. Angularly multiplexed, diffuse Fourier transform holograms of the PSRI feature space are characterized as the memory unit. Distorted input objects are correlated with the hologram, and a nonlinear phase conjugate mirror, self-pumped BaTiO3, reduces cross-correlation noise and provides object discrimination. The self-pumped phase conjugate mirror is characterized, and high diffraction efficiency bleached holograms are used in the place of thermoplastic film. Applications of the memory are also presented.